structure-screening

1. When to use, and when not

Use this skill when you need to run a bounded structural druggability screen as a pre-commitment filter (Stage 0 / Tier 0). Entry points include:

• Determining whether a target has a druggable pocket before committing it as a primary intervention target.
• Evaluating structural tractability alongside competitive landscape and mechanism-direction as a pre-commitment filter.
• Screening a target when only existing structures/models are available (no new AF3 prediction budget).

Do not use when:

• You need detailed structural characterization (that requires accepted prerequisites and a resource budget — this skill produces a fast-fail filter, not a thorough characterization).
• You need to rank different targets against each other by pocket score — the drug score ranks conformations of one target, not targets against each other (the pocket-druggability contract forbids this, and this skill preserves that constraint).
• You need to run a new AF3 prediction — retrieval of existing models only; new predictions require separate justification.
• The modality is not small-molecule or molecular-glue — pocket geometry is not the relevant tractability test for biologics, antibodies, gene therapies, etc. The screening layer will detect this and produce a not_yet_applicable assessment.

2. Relationship to pocket-druggability

pocket-druggability/SKILL.md is the interpretation authority. This skill orchestrates pocket-druggability; it does not replace or duplicate it. The relationship is:

3. Tool invocations

The structure-screen run command internally invokes oase pocket run then oase pocket analyze for each candidate structure within the screen budget. It produces oase.evidence-assessment.v1 records with relay codes and scoping preserved from the underlying pocket analysis.

Options:

• --concept-ref (required): concept reference (IC-NNN) the screen is for.
• --modality (required): intervention modality (small_molecule, molecular_glue, antibody, etc.). Non-pocket-relevant modalities produce a not_yet_applicable assessment immediately.
• --near: residues defining the intervention site (CHAIN:RESNUM, comma-separated). Passed through to oase pocket analyze --near.
• --source: structure source type (pdb, alphafold_db, existing_model). Default: pdb.
• --experimental/--no-experimental: whether structures are experimental. Auto-detected from source type if omitted.
• --max-structures: maximum structures to evaluate (default: 5).
• --max-seconds: maximum wall-clock seconds (default: 300).
• --json: machine-readable JSON output.
• --quiet: paths only.
• --claim: the claim being assessed (default: "target has a druggable binding pocket").

4. Preconditions

• Structure retrieval: existing AlphaFold DB models or PDB structures must be retrievable. If no suitable structure exists, the screen produces a not_assessed / data_unavailable record.
• fpocket on PATH: required for pocket analysis. If missing, the screen produces a tool_unavailable record.
• Screen budget: a declared budget (max structures, max wall-clock time). The screen respects this budget.

5. Workflow

Step 1: Check modality applicability

Before any structural work, check whether pocket geometry is the relevant tractability test for the declared modality.

• Small molecule / molecular glue: proceed to Step 2.
• Biologic / antibody / antisense / gene therapy / etc.: produce a not_yet_applicable assessment and stop. Do not force a pocket score onto a modality where it is not the relevant question.

Step 2: Retrieve candidate structures within budget

Retrieve existing structures for the target, prioritizing:

1. Experimental crystal structures from PDB
2. AlphaFold DB predicted models

Do not initiate a new AF3 prediction. If no suitable structure exists, produce a not_assessed / data_unavailable record with a clear note that a new prediction would require separate justification.

Step 3: Run pocket analysis per the pocket-druggability contract

For each candidate structure within the screen budget:

1. Run oase pocket run <STRUCTURE> to detect pockets.
2. Run oase pocket analyze <POCKETS_RECORD> to judge tractability.
3. If site residues are known, use --near to assess the specific intervention site.

Step 4: Assess site relevance

Do not treat ANY high-scoring cavity as a pass without assessing relevance to the intended intervention site.

• If --near was used, the verdict already answers the site question.
• If global analysis was used, note that site relevance was not confirmed — a druggable pocket elsewhere in the structure is not evidence of tractability at the intended site.

Step 5: Produce evidence assessment records

For each evaluated structure, produce a oase.evidence-assessment.v1 record that:

• Carries the evidence status mapped from the pocket verdict
• Preserves all relay codes from the pocket analysis
• Records provenance, confidence/coverage, and conformation
• Notes site relevance assessment
• Proposes the next discriminating characterization where justified

6. Mandatory relay preservation

The following relays from pocket-druggability MUST be preserved through the screening layer into the assessment record. They must not be dropped, summarized away, or contradicted by the screening layer's own verdict.

7. Evidence status mapping

Why insufficient rather than contradicted for low scores: The CDK2 calibration data (scores 0.17, 0.29, and 0.94 across three crystal structures of the same drugged ATP site) demonstrates that a sub-cutoff score on one conformation does not contradict druggability. The appropriate evidence status is insufficient — the evidence exists but is not decisive — with the fpocket.single_conformation relay carrying the calibration numbers.

8. Hard constraints

• No new AF3 predictions. Retrieval only within budget.
• No cross-target ranking by raw pocket score. The drug score ranks conformations of one target. Do not aggregate or compare scores across different targets.
• No modification of pocket-druggability. This skill wraps the existing contract; it does not change verdicts, relays, or thresholds.
• Carry relays through. If a relay fires on an underlying pocket analysis, the assessment record must preserve it.

9. What this skill produces

Following this workflow produces:

• One or more oase.evidence-assessment.v1 records per target, one per evaluated structure.
• Each record carries the evidence status, relay codes, provenance, confidence, and rationale.
• The next discriminating characterization is proposed when the evidence is insufficient.
• Budget adherence is recorded: if the screen stopped due to budget constraints, this is noted in the final assessment.